85

I know Pythonic slicing : l1[start:stop:step].

What's the use of built-in function slice?
How can I use it ?

Improve this question
1
  • 2
    this worked for me. def __slice__(self, start = None, end = None, step = None): return self.obj[slice(start, end, step)] Commented Dec 14, 2014 at 15:18

6 Answers 6

Reset to default
113

You create a slice by calling slice with the same fields you would use if doing [start:end:step] notation:

sl = slice(0,4)

To use the slice, just pass it as if it were the index into a list or string:

>>> s = "ABCDEFGHIJKL" >>> sl = slice(0,4) >>> print(s[sl]) 'ABCD'

Let's say you have a file of fixed-length text fields. You could define a list of slices to easily extract the values from each "record" in this file.

data = """\ 0010GEORGE JETSON 12345 SPACESHIP ST HOUSTON TX 0020WILE E COYOTE 312 ACME BLVD TUCSON AZ 0030FRED FLINTSTONE 246 GRANITE LANE BEDROCK CA 0040JONNY QUEST 31416 SCIENCE AVE PALO ALTO CA""".splitlines() fieldslices = [slice(*fielddef) for fielddef in [ (0,4), (4, 21), (21,42), (42,56), (56,58), ]] fields = "id name address city state".split() for rec in data: for field,sl in zip(fields, fieldslices): print("{} : {}".format(field, rec[sl])) print('') # or this same code using itemgetter, to make a function that # extracts all slices from a string into a tuple of values import operator rec_reader = operator.itemgetter(*fieldslices) for rec in data: for field, field_value in zip(fields, rec_reader(rec)): print("{} : {}".format(field, field_value)) print('')

Prints:

id : 0010 name : GEORGE JETSON address : 12345 SPACESHIP ST city : HOUSTON state : TX id : 0020 name : WILE E COYOTE address : 312 ACME BLVD city : TUCSON state : AZ id : 0030 name : FRED FLINTSTONE address : 246 GRANITE LANE city : BEDROCK state : CA id : 0040 name : JONNY QUEST address : 31416 SCIENCE AVE city : PALO ALTO state : CA
Improve this answer
Sign up to request clarification or add additional context in comments.

Comments

41

Square brackets following a sequence denote either indexing or slicing depending on what's inside the brackets:

>>> "Python rocks"[1] # index 'y' >>> "Python rocks"[1:10:2] # slice 'yhnrc'

Both of these cases are handled by the __getitem__() method of the sequence (or __setitem__() if on the left of an equals sign.) The index or slice is passed to the methods as a single argument, and the way Python does this is by converting the slice notation, (1:10:2, in this case) to a slice object: slice(1,10,2).

So if you are defining your own sequence-like class or overriding the __getitem__ or __setitem__ or __delitem__ methods of another class, you need to test the index argument to determine if it is an int or a slice, and process accordingly:

def __getitem__(self, index): if isinstance(index, int): ... # process index as an integer elif isinstance(index, slice): start, stop, step = index.indices(len(self)) # index is a slice ... # process slice else: raise TypeError("index must be int or slice")

A slice object has three attributes: start, stop and step, and one method: indices, which takes a single argument, the length of the object, and returns a 3-tuple: (start, stop, step).

Improve this answer

2 Comments

Related: stackoverflow.com/questions/2936863/…
Could anyone explain the usage of indices method? I read the documentation about one, but did get the point.
8 
>>> class sl: ... def __getitem__(self, *keys): print keys ... >>> s = sl() >>> s[1:3:5] (slice(1, 3, 5),) >>> s[1:2:3, 1, 4:5] ((slice(1, 2, 3), 1, slice(4, 5, None)),) >>>
Improve this answer

3 Comments

See Don's reply below for explanation on why this happens.
It would be beneficial to also point out that there can be multiple slices passed in to __getitem__: s[1:2:3, 1, 4:5] => (slice(1, 2, 3), 1, slice(4, 5, None)).
Might be easier to see how indices or slices inside [] are passed to special method __getitem__() without using that unpack operator * before keys.
3

The slice function returns slice objects. Slice objects are one of Python's internal types, which are optimized for read performance - all of their attributes are read-only.

Altering slice could be useful if wish to change the default behaviour. For example, lxml uses slice notation to access DOM elements (however, I haven't confirmed how they did that myself).

Improve this answer

Comments

1

While trying to answer Subset a string based on variable , I recalled that numpy has a syntactically nice way to define slice objects:

>>> import numpy as np >>> s = "The long-string instrument is a musical instrument in which the string is of such a length that the fundamental transverse wave is below what a person can hear as a tone." >>> z = np.s_[18:26] # in this case same as slice(18, 26, None) >>> s[z] 'strument'

The problem solved here is how to store the slice in a variable for later use, and np.s_ allows to do just that. Yes, it's not built-in, but as that original question was redirected here I feel like my answer belong here as well. Also, numpy was one of the reasons why so advanced slicing abilities were added to Python, IIRC.

An example of a more complex "slicing":

>>> data = np.array(range(6)).reshape((2, 3)) >>> z = np.s_[:1, 1:2] >>> data[z] array([[1]]) >>> data array([[0, 1, 2], [3, 4, 5]]) >>> z (slice(None, 1, None), slice(1, 2, None))

where z is now a tuple of slices.

Improve this answer

2 Comments

Actually in this case np.s_[18:26] just returns a Python slice object that is the same as what is constructed using the slice built-in. It can be different for other kinds of numpy-only indexing approaches.
Added, thanks! But I understood that question as how to represent slice syntactically. So this comes nearest I know of.
0

Slice objects let you programmatically generate and manipulate slices. Especially for multidimensional numpy arrays, and especially if you don't know the dimensionality in advance, you might want to construct slices on-the-fly to specify the axes or dimensions that you want.

import numpy as np dimension = np.random.randint(10) # Might have up to 10 dimensions shape = [] for d in range(dimension): shape.append(np.random.randint(10)) zz = np.random.rand(tuple(shape)) print(zz) >>> array([[0.68379351, 0.50854469, 0.64578775, 0.73441699, 0.28977396], [0.88797164, 0.81603025, 0.63978659, 0.22677299, 0.93455738], [0.0892855 , 0.28048706, 0.04262895, 0.9353467 , 0.13062249], [0.88561035, 0.93378367, 0.12124208, 0.25600301, 0.96035638]])

Here our data ended up being two dimensional (4-by-5), but there was no guarantee of that. How will you request slices from zz?

One problem is that I can't manipulate Python's slice notation. It's not valid syntax outside of a slicing operation.

my_slice = 2:3:1 >>> SyntaxError: Invalid Syntax

What if I could just build up the exact slice request I wanted in a loop, the way I can build up a string? Wouldn't that be great? I mean, sure you can use a string to do it, but it would be messy and requires eval.

your_slice_definitions = [(2,3,1), *[(None, None, None)]*(zz.ndim - 1)] my_slice_str = "" for slice_start, slice_end, slice_step in your_slice_definitions: my_slice_str += "{}:{}:{},".format(slice_start, slice_end, slice_step) eval("zz["+my_slice_str+"])

So here we are: slice objects let you do this. You can assemble lists and tuples of them on-the-fly, pass them as function parameters, sort them, shuffle them, and so on.

my_slices = [] for slice_start, slice_end, slice_step in your_slice_definitions: my_slices += slice(slice_start, slice_end, slice_step) print(zz[my_slices]) >>> array([[0.0892855 , 0.28048706, 0.04262895, 0.9353467 , 0.13062249]])
Improve this answer

Comments

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.